Temperature Sensor Faults On-Line Diagnosis Method Based on Signal Noise Analysis

A sensor fault on-line diagnosis method based on the sensor model and signal noise analysis is designed in the paper. Two fault eigenvalues, xi and xi f, are constructed for faults calibration and transient faults detection respectively. The xi r is the evolution of the ratio between two power spectrum densities (PSD) of the unfiltered noise signals while the sensor is in fault status and in reference normal status; and the xi f is the evolution of the ratio between the PSDs of the filtered noise signals. The noise signals are filtered with the soft threshold wavelet denoising method. From the xi r and xi f, the gain and the response time of the fault sensor, and the magnitude of the faults can be eatimated. In order to validate the method, several experiments are performed with a thermal parameter monitoring and control experiment platform. The results show that the method can detect the thermal conductivity degradation fault and insulation degradation fault of the temperature sensor effectively.